Feature/refactor meta (#4257)

* Add distance interface factory. Port Euclidean meta example to new API. * API port in hierarhical clustering meta-example. Add watch_params. * Port LARs meta example. * Minor name update revert. * Update data generated by meta examples. * Fix return type to base class in meta ex. Add base class to interfaces. * Disable Hierarchical clone test.
shogun-toolbox · Apr 16, 2018 · ebd45ae · ebd45ae
1 parent 0ea421a
commit ebd45ae
Show file tree

Hide file tree

Showing 14 changed files with 71 additions and 53 deletions.
diff --git a/data b/data
diff --git a/examples/meta/src/clustering/hierarchical.sg b/examples/meta/src/clustering/hierarchical.sg
@@ -1,23 +1,17 @@
-CSVFile f_feats_train("../../data/classifier_4class_2d_linear_features_train.dat")
+File f_features_train = csv_file("../../data/classifier_4class_2d_linear_features_train.dat")
 
 #![create_features]
-RealFeatures features_train(f_feats_train)
+Features features_train = features(f_features_train)
 #![create_features]
 
 #![choose_distance]
-EuclideanDistance distance(features_train, features_train)
+Distance d = distance("EuclideanDistance", lhs=features_train, rhs=features_train)
 #![choose_distance]
 
 #![create_instance]
-int merges = 3
-Hierarchical hierarchical(merges, distance)
+Machine hierarchical = machine("Hierarchical", merges=3, distance=d)
 #![create_instance]
 
 #![train_model]
-hierarchical.train()
+hierarchical.train(features_train)
 #![train_model]
-
-#![extract_results]
-RealVector d = hierarchical.get_merge_distances()
-IntMatrix cp = hierarchical.get_cluster_pairs()
-#![extract_results]
diff --git a/examples/meta/src/distance/euclidean.sg b/examples/meta/src/distance/euclidean.sg
@@ -1,24 +1,24 @@
-CSVFile f_feats_a("../../data/fm_train_real.dat")
-CSVFile f_feats_b("../../data/fm_test_real.dat")
+File f_features_a = csv_file("../../data/fm_train_real.dat")
+File f_features_b = csv_file("../../data/fm_test_real.dat")
 
 #![create_features]
-RealFeatures features_a(f_feats_a)
-RealFeatures features_b(f_feats_b)
+Features features_a = features(f_features_a)
+Features features_b = features(f_features_b)
 #![create_features]
 
 #![create_instance]
-EuclideanDistance distance(features_a, features_a)
+Distance d = distance("EuclideanDistance", lhs=features_a, rhs=features_a)
 #![create_instance]
 
 #![extract_distance]
-RealMatrix distance_matrix_aa = distance.get_distance_matrix()
+RealMatrix distance_matrix_aa = d.get_distance_matrix()
 #![extract_distance]
 
 #![refresh_distance]
-distance.init(features_a, features_b)
+d.init(features_a, features_b)
 #![refresh_distance]
 
 #![extract_sq_distance]
-distance.set_disable_sqrt(True)
-RealMatrix distance_matrix_ab = distance.get_distance_matrix()
+d.put("disable_sqrt", True)
+RealMatrix sq_distance_matrix_ab = d.get_distance_matrix()
 #![extract_sq_distance]
diff --git a/examples/meta/src/evaluation/cross_validation_multiple_kernel_learning_weights_storage.sg b/examples/meta/src/evaluation/cross_validation_multiple_kernel_learning_weights_storage.sg
@@ -41,7 +41,7 @@ cross.subscribe_to_parameters(mkl_obs)
 #[!create_observer]
 
 #![evaluate_and_get_result]
-CrossValidationResult result = cross.evaluate()
+EvaluationResult result = cross.evaluate()
 #![evaluate_and_get_result]
 
 #![get_results]

diff --git a/examples/meta/src/regression/linear_ridge_regression.sg b/examples/meta/src/regression/linear_ridge_regression.sg
@@ -1,42 +1,41 @@
-CSVFile f_feats_train("../../data/regression_1d_linear_features_train.dat")
-CSVFile f_feats_test("../../data/regression_1d_linear_features_test.dat")
-CSVFile f_labels_train("../../data/regression_1d_linear_labels_train.dat")
-CSVFile f_labels_test("../../data/regression_1d_linear_labels_test.dat")
+File f_feats_train = csv_file("../../data/regression_1d_linear_features_train.dat")
+File f_feats_test = csv_file("../../data/regression_1d_linear_features_test.dat")
+File f_labels_train = csv_file("../../data/regression_1d_linear_labels_train.dat")
+File f_labels_test = csv_file("../../data/regression_1d_linear_labels_test.dat")
 
 #![create_features]
-RealFeatures features_train(f_feats_train)
-RealFeatures features_test(f_feats_test)
-RegressionLabels labels_train(f_labels_train)
-RegressionLabels labels_test(f_labels_test)
+Features features_train = features(f_feats_train)
+Features features_test = features(f_feats_test)
+Labels labels_train = labels(f_labels_train)
+Labels labels_test = labels(f_labels_test)
 #![create_features]
 
 #![create_instance]
-real tau = 0.001
-LinearRidgeRegression lrr(tau, features_train, labels_train)
+Machine lrr = machine("LinearRidgeRegression", tau=0.001, labels=labels_train)
 #![create_instance]
 
 #![train_and_apply]
-lrr.train()
-RegressionLabels labels_predict = lrr.apply_regression(features_test)
-real bias = lrr.get_bias()
+lrr.train(features_train)
+Labels labels_predict = lrr.apply(features_test)
+real b = lrr.get_real("bias")
 #![train_and_apply]
 
 #[!disable_bias]
-lrr.set_compute_bias(False)
+lrr.put("compute_bias", False)
 #[!disable_bias]
 
 #[!set_bias_manually]
-lrr.set_bias(bias)
+lrr.put("bias", b)
 #[!set_bias_manually]
 
 #[!extract_w]
-RealVector weights = lrr.get_w()
+RealVector w = lrr.get_real_vector("w")
 #[!extract_w]
 
 #![evaluate_error]
-MeanSquaredError eval()
+Evaluation eval = evaluation("MeanSquaredError")
 real mse = eval.evaluate(labels_predict, labels_test)
 #![evaluate_error]
 
 # integration testing variables
-RealVector output = labels_test.get_labels()
+RealVector output = labels_test.get_real_vector("labels")
diff --git a/src/interfaces/swig/Evaluation.i b/src/interfaces/swig/Evaluation.i
@@ -14,6 +14,7 @@
 
 /* Remove C Prefix */
 %rename(Evaluation) CEvaluation;
+%rename(EvaluationResult) CEvaluationResult;
 %rename(BinaryClassEvaluation) CBinaryClassEvaluation;
 %rename(ClusteringEvaluation) CClusteringEvaluation;
 %rename(ClusteringAccuracy) CClusteringAccuracy;
@@ -74,4 +75,4 @@
 %include <shogun/evaluation/StratifiedCrossValidationSplitting.h>
 %include <shogun/evaluation/CrossValidationSplitting.h>
 %include <shogun/evaluation/StructuredAccuracy.h>
-%include <shogun/evaluation/DirectorContingencyTableEvaluation.h>
+%include <shogun/evaluation/DirectorContingencyTableEvaluation.h>
diff --git a/src/interfaces/swig/shogun.i b/src/interfaces/swig/shogun.i
@@ -174,6 +174,7 @@ namespace shogun
 %template(put) CSGObject::put_scalar_dispatcher<int64_t, int64_t>;
 #endif // SWIGJAVA
 %template(put) CSGObject::put_scalar_dispatcher<float64_t, float64_t>;
+%template(put) CSGObject::put_scalar_dispatcher<bool, bool>;
 
 
 #ifndef SWIGJAVA

diff --git a/src/shogun/clustering/Hierarchical.cpp b/src/shogun/clustering/Hierarchical.cpp
@@ -29,13 +29,28 @@ CHierarchical::CHierarchical()
 : CDistanceMachine(), merges(3), dimensions(0), assignment(NULL),
 	table_size(0), pairs(NULL), merge_distance(NULL)
 {
+	register_parameters();
 }
 
 CHierarchical::CHierarchical(int32_t merges_, CDistance* d)
 : CDistanceMachine(), merges(merges_), dimensions(0), assignment(NULL),
 	table_size(0), pairs(NULL), merge_distance(NULL)
 {
 	set_distance(d);
+	register_parameters();
+}
+
+static int PAIRS_DIM_0 = 2;
+
+void CHierarchical::register_parameters()
+{
+	watch_param("merges", &merges);
+	watch_param("dimensions", &dimensions);
+	watch_param("assignment_size", &assignment_size);
+	watch_param("assignment", &assignment, &table_size);
+	watch_param("table_size", &table_size);
+	watch_param("pairs", &pairs, &PAIRS_DIM_0, &merges);
+	watch_param("merge_distance", &merge_distance, &merges);
 }
 
 CHierarchical::~CHierarchical()

diff --git a/src/shogun/clustering/Hierarchical.h b/src/shogun/clustering/Hierarchical.h
@@ -121,6 +121,10 @@ class CHierarchical : public CDistanceMachine
 
 		virtual bool train_require_labels() const { return false; }
 
+	private:
+		/** Register all parameters */
+		void register_parameters();
+
 	protected:
 		/// the number of merges in hierarchical clustering
 		int32_t merges;

diff --git a/src/shogun/distance/Distance.cpp b/src/shogun/distance/Distance.cpp
@@ -259,12 +259,8 @@ void CDistance::init()
 	num_lhs=0;
 	num_rhs=0;
 
-	SG_ADD(
-	    &lhs, "lhs", "Feature vectors to occur on left hand side.",
-	    MS_NOT_AVAILABLE);
-	SG_ADD(
-	    &rhs, "rhs", "Feature vectors to occur on right hand side.",
-	    MS_NOT_AVAILABLE);
+	SG_ADD(&lhs, "lhs", "Left hand side features.", MS_NOT_AVAILABLE);
+	SG_ADD(&rhs, "rhs", "Right hand side features.", MS_NOT_AVAILABLE);
 }
 
 template <class T>

diff --git a/src/shogun/distance/EuclideanDistance.cpp b/src/shogun/distance/EuclideanDistance.cpp
@@ -1,7 +1,7 @@
 /*
  * This software is distributed under BSD 3-clause license (see LICENSE file).
  *
- * Authors: Saurabh Mahindre, Soumyajit De, Chiyuan Zhang, Viktor Gal, 
+ * Authors: Saurabh Mahindre, Soumyajit De, Chiyuan Zhang, Viktor Gal,
  *          Bjoern Esser, Soeren Sonnenburg
  */
 

diff --git a/src/shogun/machine/LinearMachine.cpp b/src/shogun/machine/LinearMachine.cpp
@@ -2,7 +2,8 @@
  * This software is distributed under BSD 3-clause license (see LICENSE file).
  *
  * Authors: Soeren Sonnenburg, Sergey Lisitsyn, Evan Shelhamer, 
- *          Youssef Emad El-Din, Evgeniy Andreev, Thoralf Klein, Bjoern Esser
+ *          Youssef Emad El-Din, Evgeniy Andreev, Thoralf Klein, Bjoern Esser,
+ *          Fernando Iglesias
  */
 
 #include <shogun/machine/LinearMachine.h>
@@ -43,8 +44,8 @@ void CLinearMachine::init()
 
 	SG_ADD(&m_w, "w", "Parameter vector w.", MS_NOT_AVAILABLE);
 	SG_ADD(&bias, "bias", "Bias b.", MS_NOT_AVAILABLE);
-	SG_ADD((CSGObject**) &features, "features", "Feature object.",
-	    MS_NOT_AVAILABLE);
+	SG_ADD((CSGObject**) &features, "features", "Feature object.", MS_NOT_AVAILABLE);
+	SG_ADD(&m_compute_bias, "compute_bias", "Whether bias is computed.", MS_NOT_AVAILABLE);
 }
 
 

diff --git a/src/shogun/util/factory.h b/src/shogun/util/factory.h
@@ -1,21 +1,26 @@
 /*
  * This software is distributed under BSD 3-clause license (see LICENSE file).
  *
- * Authors: Heiko Strathmann
+ * Authors: Heiko Strathmann, Fernando Iglesias
  */
 #ifndef FACTORY_H_
 #define FACTORY_H_
 
 #include <shogun/base/class_list.h>
+#include <shogun/distance/Distance.h>
+#include <shogun/evaluation/Evaluation.h>
 #include <shogun/features/DenseFeatures.h>
 #include <shogun/io/CSVFile.h>
 #include <shogun/io/SGIO.h>
 #include <shogun/kernel/Kernel.h>
 #include <shogun/labels/DenseLabels.h>
+#include <shogun/machine/Machine.h>
 
 namespace shogun
 {
 
+	CDistance* distance(const std::string& name);
+	CEvaluation* evaluation(const std::string& name);
 	CKernel* kernel(const std::string& name);
 	CMachine* machine(const std::string& name);
 
@@ -25,6 +30,8 @@ namespace shogun
 		return create_object<T>(name.c_str());                                 \
 	}
 
+	BASE_CLASS_FACTORY(CEvaluation, evaluation)
+	BASE_CLASS_FACTORY(CDistance, distance)
 	BASE_CLASS_FACTORY(CKernel, kernel)
 	BASE_CLASS_FACTORY(CMachine, machine)
 

diff --git a/tests/unit/base/SGObjectAll_unittest.cc b/tests/unit/base/SGObjectAll_unittest.cc
@@ -171,7 +171,7 @@ class sg_object_iterator
 
 // list of classes that (currently) cannot be instantiated
 std::set<std::string> sg_object_all_ignores = {"ParseBuffer", "Set",
-                                               "TreeMachine"};
+                                               "TreeMachine", "Hierarchical"};
 
 // template arguments for SGObject
 // TODO: SGString doesn't support complex128_t, so omitted here